Skip to main content
Version: Dev

Compatibility and authorities

What ObjectStoreViewer has actually been checked against, and which upstream source decides each question.

Format authorities​

QuestionAuthority
Barman Cloud layout, metadata, and compressionBarman source, documentation, and pinned generated repositories
WAL naming and segment arithmeticPostgreSQL source
pgBackRest layout, info files, and reference chainspgBackRest source, release policy, its documented stable JSON output, and pinned generated repositories
CNPG object-store topologyCloudNativePG and the Barman Cloud Plugin documentation

The format behavior is reimplemented in Go. Implementation code is not copied from Barman or pgBackRest, which keeps this project's Apache-2.0 licensing intact.

Verified versions​

ComponentChecked against
Barman3.19.1 — a genuine barman-cloud-backup and WAL archive journey runs inside make test-integration
PostgreSQL WAL grammar18 (src/include/access/xlog_internal.h): 24-hex-digit names, 1 MiB – 1 GiB power-of-two segments, 2^32 / wal_segment_size segments per log ID, .partial, timeline .history, backup-history names
Barman WAL grammar and hash directory3.19.1 (barman/xlog.py)
Barman compression suffixes3.19.1 (barman/compression.py): .gz, .bz2, .xz, .snappy, .zst, .lz4
Go toolchain1.26.5
Object storesS3/MinIO, Azure/Azurite, GCS/fake-gcs-server — all pinned by digest
Architectureslinux/amd64, linux/arm64
Restricted runtimedistroless non-root; arbitrary UID/GID, read-only root, dropped capabilities, no-new-privileges
OpenShift4.20 under restricted-v2 (live sidecar Pod and channel mechanics, 2026-07-29)

Arithmetic detail worth knowing: lowercase hexadecimal WAL input is accepted, as Barman does, while rendered range boundaries are normalized to uppercase.

Compression support​

Barman Cloud distinguishes base-backup from WAL compression, and so does the viewer:

ObjectSupported compression
base backupuncompressed, gzip, bzip2, snappy, lz4
WALuncompressed, gzip, bzip2, snappy, lz4, xz, zstd

A suffix is normalized only after the underlying WAL/history name and its hash-directory placement validate. Anything else is bounded unknown diagnostic evidence, not a guess.

CloudNativePG basis​

Checked against the Barman Cloud Plugin 0.13.0 documentation and Barman 3.19:

  • the plugin requires CloudNativePG 1.26 or later and recommends 1.27 or later;
  • it runs the Barman Cloud client in a database-pod sidecar; and
  • base backups are tarballs under the Barman Cloud base layout.

Both CNPG integrations are in scope: the legacy in-tree barmanObjectStore configuration and the Barman Cloud Plugin's ObjectStore CRD in API group barmancloud.cnpg.io.

The plugin keeps ObjectStore.serverName empty

A server name can instead be selected through Cluster plugin parameters, so an Kubernetes operator composing this viewer must pass the effective server names or rely on validated discovery.

Explicitly out of scope​

Barman 3.19 supports block-level incremental backups streamed to cloud through a configured Barman server. That is not the Barman Cloud client topology the CNPG plugin exposes, so such a repository is reported as incompatible_format/unknown — never as ordinary full-backup evidence.

This is not a permanent exclusion. If an accepted CNPG or CNPG-I integration exposes Barman server streaming incrementals, support becomes a release gate: format-owned parent/chain fields, dependency validation, fixtures, and failure-to-unknown tests must land before that CNPG/Plugin/Barman tuple is called supported.

Also outside the current profile: Kubernetes/CSI VolumeSnapshot evidence (a consumer-side source), pgBackRest interpretation, multiple repositories or scopes in one process, and raw downloads.

Discovery validation​

When a scope list is empty, discovery is delegated to the format — and it validates rather than guesses:

  • Barman accepts only prefixes with a recognizable base/ or wals/ subtree.
  • pgBackRest accepts only stanzas represented consistently under backup/ or archive/.

Invalid candidates are ignored and reported as diagnostics. Names are trimmed, deduplicated, and sorted deterministically.

Upstream references​

Barman

CloudNativePG

pgBackRest

PostgreSQL